Do Newer Models Hold Up as Context Fills? Reading time ~20 minutes

My experience has always been that AI models get pretty dumb after about 50% of their context window is filled. @vikhyatk on x.com challenged my assumption and said it’s been getting better. I decided to put on a little test. The below experiment was fully conducted and documented by gpt-5.2-codex......

Sharing Skills: Claude Marketplace to VS Code Workspaces Reading time ~4 minutes

Part 3 of 3 in the Skills Catalog series (Part 1 Part 2) You’ve built a skills catalog. Nested domains. Scripts that execute on demand. Now the final question: How do you share this across your team? Anthropic distributes skills through their marketplace and Skills API—portable, reusable knowledge packages. GitHub......

Learning from Anthropic: Nested Skills and Scripts Reading time ~7 minutes

Part 2 of 3 in the Skills Catalog series (Part 1 Part 3) Part 1 covered the catalog pattern. Now: how do you organize knowledge that scales to hundreds of skills? Anthropic’s skills repository shows patterns for complex domains: Nested structures for complex domains Scripts as executable knowledge Supporting resources......

Bringing Anthropic Skills to GitHub Copilot Reading time ~5 minutes

Part 1 of 3 in the Skills Catalog series (Part 2 Part 3) Anthropic just announced Skills—modular, discoverable knowledge for Claude. Each skill is a SKILL.md file with metadata. Claude discovers them through the description field. No central catalog, no manual wiring. Can we bring this to GitHub Copilot? Yes—with......

Why Use AI CLIs When GUIs Exist? Reading time ~1 minute

People keep asking me why they’d ever use an AI command-line tool when the GUI works fine. The answer: CLIs let you mix deterministic scripting with non-deterministic AI judgment calls. Pure AI is hard to steer. Pure scripting breaks on edge cases. CLI tools blend both. The Difference in Action......